Screen Builder instructions
Stocks are selected from a specified database by applying a
series of rules. Each rule is a filter that excludes stocks that
don't pass - a stock must pass all rules to be selected.
Rules are generally entered as algebriac/programming expressions
of data fields, functions, and numbers (and ocasionally quoted strings),
with the comparison operators (= != > < >= <=),
arithmetic operators (+ - * / **), and parentheses.
The boolean operators "&" (AND) and "|"
(OR) can be used to satisfy multiple conditions in a single expression.
The field (column) names are represented by alphanumeric codes, as
explained in this field guide.
Technical functions are independent of any database, and are based
solely on stock data (price, etc.). Functions take one or more
parameters, usually beginning with a lookback period expressed in days.
Available functions are:
rs(days):
traditional relative strength
ema(days):
exponential moving average
rsi(days):
relative strength index
Functions may also have an additional parameter that specifies a
number of days to lag the technical measurement used.
Sort rules are entered using the keywords "top" (i.e.
descending) and "bottom" (i.e. ascending), followed by
a number. Tie-breakers can be included as a comma-separated list of
expressions before the "top" or "bottom".
You may select a percentage of the current stock list (rounding down),
instead of a fixed number of stocks, by appending a "%" to
the number.
Some screens are built with fields that don't exist during the entire
backtest period. To extend the backtest further in these cases, you can
assign replacement values to missing fields. Use the special form
"(fld1 | fld2)" to use the field fld2 when
fld1 is unavailable. Similarly, you can use a constant when
a field is unavailable, e.g. "(fld1 | 0)". Many screens also
use total-return fields not available from earlier VL data (such as
tr4w - Total Return 4-Week), but this is handled
automatically from the daily price data.
All this may sound pretty complicated. It is, but not unreasonably so.
For example, here's the
Small_Value
screen. If you can make sense of that, you should be able to build your
own screens as well.
|